Research And Implementation Of Cross Network User Representation Under Heterogeneous Information

Complicated social networks have brought a variety of network services to people’s lives.The user association mining between different social networks has become a hot research topic at present,which has a wide range of application values in the recommendation system,false account detection,user portrait and other aspects.Network user representation technology is the key technology of cross-platform association,but it still faces the following challenges:heterogeneity of structure,incomplete relation extraction,complex attribute information and weak attribute representation ability.Aiming at the problem of user representation of heterogeneous information fusion in cross-social network environment,this paper studies and implements a cross-network user representation technology for heterogeneous networks.Firstly,in view of the complexity and heterogeneity of heterogeneous information networks,a network representation technology based on improved random walk algorithm was proposed.Network structure optimization was used to solve the problem of network topology inconsistence from two aspects of network node filtering and network structure completion.Deep traversal embedding and breadth traversal embedding were used to extract structural information from global and local perspectives.To realize cross-network user representation based on network structure;Secondly,in view of the diversity and heterogeneity of attributes,a network representation technology based on multi-text level embedding is proposed.For the short text content of user name and address information,the way of character level and word level embedding is adopted.For the text content as long as the user’s document,the long text embedding including topic analysis and semantic representation is used.To realize cross-network user representation based on user attributes;Aiming at the problem of insufficient information of fusion network structure and user attributes,the graph attention network aggregation is finally used to achieve the crossnetwork user representation of fusion structure and attribute characteristics.Experiments on real social network data sets show that the crossnetwork user representation technique proposed in this paper has good experimental results in user identity alignment tasks.Finally,a cross-network user alignment system is developed by using the above technology.The system realizes the effective representation of cross-network users and the identification of users with the same identity,and the effectiveness of the system is verified by tests.